DISCRIMINATION AND GENERALIZATION

The decade of the 1990s witnessed acceleration in the convergence of theoretical and experimental studies of discrimination and generalization from the domains of classical conditioning and instrumental (operant) learning. Classical conditioning refers to the establishment of behavioral adaptations (conditioned responses; CRs) by the methods of Pavlov. Instrumental learning is a general term for goal-seeking behavior, and operant conditioning refers to reinforcement learning by the methods of Skinner. The term discrimination refers to the capacity of organisms to learn different modes of behavior depending on signals or cues from the environment about the imminence or accessibility of reinforcement. Generalization refers to stimulus generalization, the capacity for signals or cues that are different from those used for establishing learned behavior to evoke this behavior. Stimulus generalization in classical conditioning refers to the capacity of a stimulus other than the conditioned stimulus to evoke a CR. In operant conditioning, one set of stimuli, an occasion setter (OS), might evoke the behavior controlled by another OS, depending on their shared features or similarity. Skinner coined the term occasion setter to refer to signals or cues that predict reinforcement. In recent years, the term occasion setting has been extended to encompass both classical and operant forms of behavioral learning. As a consequence of this mixing, the terminology and paradigms used in occasion setting research borrow from the two domains. The mixture of the two domains has led to a healthy integration of methods and ideas (Schmajuk and Holland, 1998).

The convergence of ideas about discriminations and generalization from classical and operant conditioning began during the late 1960s, when the principles of stimulus control enunciated by operant-conditioning studies involving pigeons were found to extend to eyeblink conditioning in rabbits (Moore, 1972). Specifically, generalization along an auditory frequency dimension shares many of the same characteristics as visual wavelength generalization in pigeons. Conditioned stimulus preexposure (latent inhibition) also generalizes along an auditory frequency dimension, with gradients forming an inverted V shape (Siegel, 1972). Latent inhibition refers to retarded classical conditioning as a result of preexposure to the CS. Inverted-V-shape generalization gradients have also been observed with tonal stimuli trained with Pavlov's conditioned inhibition procedure (Mis, Lumia, and Moore, 1972). Inverted-V generalization gradients have been observed in pigeon operant tasks (Hearst, 1969).

Paradigms for Occasion Setting

Occasion setters are stimulus features, such as the presence of a light or tone, that serve as discriminative stimuli. For example, the presence of a light might signal that operant responses will be reinforced. The absence of this light would signal that operant responses are not reinforced. In general, a feature-positive paradigm is one in which the OS signals reinforcement; a feature-negative paradigm is one in which the OS signals the absence of reinforcement. The presence of a light or tone might signal reinforcement, whereas its absence signals nonreinforcement. In classical conditioning a feature-positive occasion-setting task would involve adding a feature to the CS. For example, if the CS is a tone, the addition of a light sets the occasion for reinforcement, whereas the tone alone would not signal reinforcement. A feature-negative task would be one in which the light, instead of signaling reinforcement when presented with the tone, would signal its absences. If the occasion setter overlaps the CS and signals the reinforcement, it can result in a more robust CR than would otherwise be the case. By contrast, if OS overlaps the CS and signals the absence of the US, it can inhibit the CR. This feature-negative discrimination recognizing the absence of an OS as a signal of the absence of the US is a relatively difficult discrimination for animals to master. In fact, a classical conditioning situation like this is called a conditioned inhibition paradigm, and research has shown that this is relatively difficult discrimination for animals to master.

Pattern Learning

The relative difficulty in learning feature-negative discrimination as compared with feature-positive discrimination learning extends also to the paradigms of positive and negative patterning. Demonstrations of patterning in both classical and operant conditioning tasks involve two signaling stimuli, A and B, which could be occasion setters, and three types of trials. In positive patterning, the trial types are AB+, A-, and B-. Animals readily learn to respond more vigorously to the reinforced stimulus compound, AB, and to inhibit responses to A- and B-, where (+) denotes a reinforced trial and (-) denotes a nonreinforced trial. In negative patterning, the trial types are AB-, A+, and B+. In order to behave appropriately in a negative-patterning task, the animal must somehow learn that responding to the stimulus compound AB must be suppressed. This can happen only if the compound stimulus has unique features that are subsets of neither A nor B (Pearce and Bouton, 2001). That is, the compound has a configural component that is shared by neither A nor B.

The general difficulty in learning a negative-patterning discrimination suggests that this configural component is often overshadowed by the nonconfigural aspects of A and B, each of which in isolation drives a tendency to respond by virtue of their association with reinforcement. At the same time, some negative-patterning tasks can be mastered comparatively easily, suggesting that the compound AB forms a unique pattern that, although similar to A and B, nevertheless is treated as a whole, thereby allowing the animal to master the discrimination.

Perceptual Learning

Exposure to discriminative stimuli enhances subsequent discrimination learning, an effect known as perceptual learning (Hall, 1991). The phenomenon of perceptual learning might seem to contradict the idea that exposure to stimuli retards later learning, as in latent inhibition. But latent inhibition can actually account for perceptual learning, if we assume that pre-exposed discriminatory stimuli share features in common. Latent inhibition develops to shared and unshared features alike, but the common features lose associability more rapidly than the unshared features. Animals learn to ignore the shared features, allowing the unshared features to dominate subsequent discrimination learning and thereby facilitating performance (McLaren and Mackintosh, 2000).

Not only do the shared features undergo more latent inhibition than the unshared features, the unshared features also can become mutually inhibitory through mechanisms of Pavlovian conditioned inhibition, provided the two stimuli are alternated during the preexposure phase (Dwyer, Bennett, and Mackintosh, 2001). The unique aspect of one preexposed stimulus can actively suppress associations with the unique aspect of the other preexposed stimulus, and vice versa. Thus, any tendency for one stimulus to elicit a representation of the other is reduced, and this promotes perceptual learning.

Acquired Distinctiveness

Acquired distinctiveness refers to enhanced discrimination learning to stimuli or stimulus dimensions that had been used in a prior discrimination task. Unlike perceptual learning, however, acquired distinctiveness appears to entail more than latent inhibition of shared stimulus features and mutual inhibition of unique features. The additional ingredient is correlation with reinforcement. The nature of this correlation remains in question. George and Pearce (1999) have argued that acquired distinctiveness stems from prior learning about the relevance of the stimuli for solving problems based on the same class of reinforcers. The relevance-to-reinforcement account of acquired distinctiveness does not imply that prior learning about the lack of correlation of the stimuli for reinforcement impedes discrimination learning.

Time Discrimination

Temporal control of behavior is one of the signature features of operant conditioning methods, and instances of temporal discrimination and generalization for duration has been well documented in operant conditioning tasks in animals and humans (e.g., Wearden and Bray, 2001). Similar instances of temporal control of behavior occur in classical conditioning (Gallistel and Gibbon, 2000; Kehoe and Macrae, 2002). Conditioned-response timing depends on the CS-US interval(s) used in training. Conditioned responses are timed so that they achieve maximal amplitude at the anticipated time(s) of the US, and this occurs whether the CS is defined operationally as the onset or the offset of a stimulus (e.g., Kehoe, Schreurs, Macrae, and Gormezano, 1995). In classical conditioning the timing of conditioned responses becomes more variable as the CS-US interval increases (Gallistel and Gibbon, 2000; White, Kehoe, Choi, and Moore, 2000). Temporal discriminative control of classically conditioned responses also occur in occasion setting (Kehoe, Palmer, Weiderman, and Macrae, 2000; see Figure 1).

See also:CLASSICAL CONDITIONING: BEHAVIORAL PHENOMENA

Bibliography

Dwyer, D. W., Bennett, C. H., and Mackintosh, N. J. (2001). Evidence for inhibitory associations between the unique elements of two compound flavors. Quarterly Journal of Experimental Psychology 54B, 97-107.

Gallistel, C. R., and Gibbon, J. (2000). Time, rate, and conditioning. Psychological Review 107, 289-344.

George, D. N., and Pearce, J. M. (1999). Acquired distinctiveness is controlled by stimulus relevance and not correlation with reward. Journal of Experimental Psychology: Animal Behavior Processes 25, 363-373.

Hall, G. (1991). Perceptual and associative learning. Oxford: Clarendon Press.

Hearst, E. (1969). Excitation, inhibition, and discrimination learning. In N. J. Mackintosh and W. K. Honig, eds., Fundamental issues in associative learning, pp. 1-41. Halifax: Dalhousie University Press.

Kehoe, E. J., and Macrae, M. (2002). Fundamental behavioral and findings in classical conditioning. In J. W. Moore, ed., A neuroscientist's guide to classical conditioning, pp. 171-231. New York: Springer.

Kehoe, E. J., Palmer, N., Weiderman, G., and Macrae, M. (2000). The effect of feature target intervals in conditioned discriminations on acquisition and expression of conditioned nictitating membrane and heart-rate responses in the rabbit. Animal Learning and Behavior 28, 80-91.

Kehoe, E. J., Schreurs, B. G., Macrae, M, and Gormezano, I. (1995). Effects of modulating tone frequency, intensity, and duration on the classically conditioned nictitating membrane response. Psychobiology 23, 103-115.

McLaren, I. P. L., and Mackintosh, N. J. (2000). An elemental model of associative learning: Latent inhibition and perceptual learning. Animal Learning and Behavior 28, 211-246.

Mis, F. W., Lumia, A. W., and Moore, J. W. (1972). Inhibitory control of the classically conditioned nictitating membrane response of the rabbit. Behavior Research Methods and Instrumentation 4, 297-299.

Moore, J. W. (1972). Stimulus control: Studies of auditory generalization in rabbits. In A. H. Black and W. P. Prokasy, eds., Classical conditioning II: Current research and theory, pp. 206-230. New York: Appleton-Century-Crofts.

Pearce, J. M., and Bouton, M. E. (2001). Theories of associative learning in animals. Annual Review of Psychology 52, 111-139.

Schmajuk, N. A., and Holland, P. C., eds. (1998). Occasion setting. Washington, DC: American Psychological Association.

Siegel, S. (1972). Latent inhibition and eyelid conditioning. In A. H. Black and W. P. Prokasy, eds., Classical conditioning II: Current research and theory, pp. 231-247. New York: Appleton-Century-Crofts.

Weardon, J. H., and Bray, S. (2001). Scalar timing without reference memory? Episodic temporal generalization and bisection in humans. Quarterly Journal of Experimental Psychology 54B, 289-309.

White, N. E., Kehoe, E. J., Choi, J-S, and Moore, J. W. (2000). Coefficient of variation in timing of the classically conditioned eye-blink in rabbits. Animal Learning and Behavior 28, 520-524.

David R.Thomas

Revised byJohn W.Moore